NewCategorization.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 208
    },
    "colab_type": "code",
    "id": "CO-mVCv5ZgP0",
    "outputId": "351cc027-9ea4-4174-d857-036295333ae1"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Requirement already satisfied: kaggle in /usr/local/lib/python3.6/dist-packages (1.5.6)\n",
      "Requirement already satisfied: tqdm in /usr/local/lib/python3.6/dist-packages (from kaggle) (4.38.0)\n",
      "Requirement already satisfied: python-dateutil in /usr/local/lib/python3.6/dist-packages (from kaggle) (2.8.1)\n",
      "Requirement already satisfied: urllib3<1.25,>=1.21.1 in /usr/local/lib/python3.6/dist-packages (from kaggle) (1.24.3)\n",
      "Requirement already satisfied: requests in /usr/local/lib/python3.6/dist-packages (from kaggle) (2.21.0)\n",
      "Requirement already satisfied: six>=1.10 in /usr/local/lib/python3.6/dist-packages (from kaggle) (1.12.0)\n",
      "Requirement already satisfied: python-slugify in /usr/local/lib/python3.6/dist-packages (from kaggle) (4.0.0)\n",
      "Requirement already satisfied: certifi in /usr/local/lib/python3.6/dist-packages (from kaggle) (2020.4.5.1)\n",
      "Requirement already satisfied: idna<2.9,>=2.5 in /usr/local/lib/python3.6/dist-packages (from requests->kaggle) (2.8)\n",
      "Requirement already satisfied: chardet<3.1.0,>=3.0.2 in /usr/local/lib/python3.6/dist-packages (from requests->kaggle) (3.0.4)\n",
      "Requirement already satisfied: text-unidecode>=1.3 in /usr/local/lib/python3.6/dist-packages (from python-slugify->kaggle) (1.3)\n"
     ]
    }
   ],
   "source": [
    "!pip install kaggle"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "Ej6JX8vidnsL"
   },
   "outputs": [],
   "source": [
    "!mkdir ~/.kaggle"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "6TCtfiJndyGz"
   },
   "outputs": [],
   "source": [
    "!echo '{\"username\":\"<you kaggle user_name>\",\"key\":\"<your kaggle access_key>\"}' > ~/.kaggle/kaggle.json"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "nYHOc34eewZG"
   },
   "outputs": [],
   "source": [
    "!chmod 600 ~/.kaggle/kaggle.json"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 69
    },
    "colab_type": "code",
    "id": "9mGx71hnfEcZ",
    "outputId": "dfb03865-c838-4042-8655-668afc5a6a6e"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Downloading news-category-dataset.zip to /content\n",
      " 35% 9.00M/25.4M [00:00<00:01, 14.7MB/s]\n",
      "100% 25.4M/25.4M [00:00<00:00, 40.1MB/s]\n"
     ]
    }
   ],
   "source": [
    "!kaggle datasets download -d rmisra/news-category-dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 34
    },
    "colab_type": "code",
    "id": "n-VXO48gfrfp",
    "outputId": "20843ae1-6998-4cd8-b348-f187970aa415"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "news-category-dataset.zip  sample_data\n"
     ]
    }
   ],
   "source": [
    "!ls"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 52
    },
    "colab_type": "code",
    "id": "qIWeqH3Qgjjw",
    "outputId": "9f35c0ff-1e72-4341-8d36-42b10caaeac5"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Archive:  news-category-dataset.zip\n",
      "  inflating: News_Category_Dataset_v2.json  \n"
     ]
    }
   ],
   "source": [
    "!unzip news-category-dataset.zip"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 34
    },
    "colab_type": "code",
    "id": "1DHdwuYTgoP3",
    "outputId": "97a878f1-c87b-4fd6-80a7-b879dd17aa36"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "News_Category_Dataset_v2.json  news-category-dataset.zip  sample_data\n"
     ]
    }
   ],
   "source": [
    "!ls"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 69
    },
    "colab_type": "code",
    "id": "SmC-0um18nFL",
    "outputId": "1b255de8-9de1-4ca0-97d1-af1fddb8ceef"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[nltk_data] Downloading package stopwords to /root/nltk_data...\n",
      "[nltk_data]   Unzipping corpora/stopwords.zip.\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "True"
      ]
     },
     "execution_count": 10,
     "metadata": {
      "tags": []
     },
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import nltk\n",
    "nltk.download(\"stopwords\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "ScXe789ogqw8"
   },
   "outputs": [],
   "source": [
    "# lets try to build a model to classify the news article into its group"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 34
    },
    "colab_type": "code",
    "id": "-7DJ4vVOhBWH",
    "outputId": "7118aec4-635d-4e25-b133-702701d6fc7f"
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "import json\n",
    "import re\n",
    "from nltk.corpus import stopwords\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "import matplotlib.pyplot as plt\n",
    "%matplotlib inline\n",
    "\n",
    "from keras.preprocessing.text import Tokenizer\n",
    "from keras.preprocessing.sequence import pad_sequences\n",
    "from keras.utils import to_categorical\n",
    "from keras import regularizers\n",
    "\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.preprocessing import LabelEncoder\n",
    "from sklearn.metrics import accuracy_score, confusion_matrix, classification_report\n",
    "from sklearn.feature_extraction.text import CountVectorizer, TfidfTransformer\n",
    "from sklearn.naive_bayes import MultinomialNB\n",
    "from sklearn.pipeline import Pipeline\n",
    "from sklearn.linear_model import SGDClassifier\n",
    "\n",
    "\n",
    "from keras.models import Sequential\n",
    "from keras.layers import Embedding\n",
    "from keras.layers import Flatten, Dense, Dropout\n",
    "from keras.layers import SimpleRNN\n",
    "from keras.layers import LSTM\n",
    "\n",
    "pd.set_option('display.max_colwidth', 200)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "Ks2eOpjjhSa-"
   },
   "outputs": [],
   "source": [
    "# reading the news dataset which is in json format"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "pnDqejmkhX3w"
   },
   "outputs": [],
   "source": [
    "news_text = []\n",
    "news_label = []\n",
    "\n",
    "with open('News_Category_Dataset_v2.json') as file:\n",
    "    for news in file:\n",
    "        j_news = json.loads(news)\n",
    "        news_label.append(j_news['category'])\n",
    "        news_text.append(j_news['headline'] + ' ' + j_news['short_description'])        "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 34
    },
    "colab_type": "code",
    "id": "wfmyP-ZLhbLR",
    "outputId": "04a4ca14-be6a-4d5c-97ce-24eb58b6c436"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(200853, 200853)"
      ]
     },
     "execution_count": 15,
     "metadata": {
      "tags": []
     },
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(news_label), len(news_text)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "f2wN52KNhn7A"
   },
   "outputs": [],
   "source": [
    "# we have 200853 news articles with their categories tagged."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "o2_o-j4bhrxZ"
   },
   "outputs": [],
   "source": [
    "news_df = pd.DataFrame(data=list(zip(news_text, news_label)), columns=['text', 'cat'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 258
    },
    "colab_type": "code",
    "id": "AHvlsACHhuxw",
    "outputId": "d836022c-49a5-40ed-dbe3-7c34ceecac55"
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>text</th>\n",
       "      <th>cat</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>There Were 2 Mass Shootings In Texas Last Week, But Only 1 On TV She left her husband. He killed their children. Just another day in America.</td>\n",
       "      <td>CRIME</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>Will Smith Joins Diplo And Nicky Jam For The 2018 World Cup's Official Song Of course it has a song.</td>\n",
       "      <td>ENTERTAINMENT</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>Hugh Grant Marries For The First Time At Age 57 The actor and his longtime girlfriend Anna Eberstein tied the knot in a civil ceremony.</td>\n",
       "      <td>ENTERTAINMENT</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>Jim Carrey Blasts 'Castrato' Adam Schiff And Democrats In New Artwork The actor gives Dems an ass-kicking for not fighting hard enough against Donald Trump.</td>\n",
       "      <td>ENTERTAINMENT</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>Julianna Margulies Uses Donald Trump Poop Bags To Pick Up After Her Dog The \"Dietland\" actress said using the bags is a \"really cathartic, therapeutic moment.\"</td>\n",
       "      <td>ENTERTAINMENT</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                                                                                                                                              text            cat\n",
       "0                    There Were 2 Mass Shootings In Texas Last Week, But Only 1 On TV She left her husband. He killed their children. Just another day in America.          CRIME\n",
       "1                                                             Will Smith Joins Diplo And Nicky Jam For The 2018 World Cup's Official Song Of course it has a song.  ENTERTAINMENT\n",
       "2                          Hugh Grant Marries For The First Time At Age 57 The actor and his longtime girlfriend Anna Eberstein tied the knot in a civil ceremony.  ENTERTAINMENT\n",
       "3     Jim Carrey Blasts 'Castrato' Adam Schiff And Democrats In New Artwork The actor gives Dems an ass-kicking for not fighting hard enough against Donald Trump.  ENTERTAINMENT\n",
       "4  Julianna Margulies Uses Donald Trump Poop Bags To Pick Up After Her Dog The \"Dietland\" actress said using the bags is a \"really cathartic, therapeutic moment.\"  ENTERTAINMENT"
      ]
     },
     "execution_count": 19,
     "metadata": {
      "tags": []
     },
     "output_type": "execute_result"
    }
   ],
   "source": [
    "news_df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "Ns8DOYKQhxUw"
   },
   "outputs": [],
   "source": [
    "# lets have a look at the news categories and their distribution"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 746
    },
    "colab_type": "code",
    "id": "HrX6iqpjh5-u",
    "outputId": "3c24f893-c085-4f43-cdf8-8f3df03e0d25"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "POLITICS          32739\n",
       "WELLNESS          17827\n",
       "ENTERTAINMENT     16058\n",
       "TRAVEL             9887\n",
       "STYLE & BEAUTY     9649\n",
       "PARENTING          8677\n",
       "HEALTHY LIVING     6694\n",
       "QUEER VOICES       6314\n",
       "FOOD & DRINK       6226\n",
       "BUSINESS           5937\n",
       "COMEDY             5175\n",
       "SPORTS             4884\n",
       "BLACK VOICES       4528\n",
       "HOME & LIVING      4195\n",
       "PARENTS            3955\n",
       "THE WORLDPOST      3664\n",
       "WEDDINGS           3651\n",
       "WOMEN              3490\n",
       "IMPACT             3459\n",
       "DIVORCE            3426\n",
       "CRIME              3405\n",
       "MEDIA              2815\n",
       "WEIRD NEWS         2670\n",
       "GREEN              2622\n",
       "WORLDPOST          2579\n",
       "RELIGION           2556\n",
       "STYLE              2254\n",
       "SCIENCE            2178\n",
       "WORLD NEWS         2177\n",
       "TASTE              2096\n",
       "TECH               2082\n",
       "MONEY              1707\n",
       "ARTS               1509\n",
       "FIFTY              1401\n",
       "GOOD NEWS          1398\n",
       "ARTS & CULTURE     1339\n",
       "ENVIRONMENT        1323\n",
       "COLLEGE            1144\n",
       "LATINO VOICES      1129\n",
       "CULTURE & ARTS     1030\n",
       "EDUCATION          1004\n",
       "Name: cat, dtype: int64"
      ]
     },
     "execution_count": 21,
     "metadata": {
      "tags": []
     },
     "output_type": "execute_result"
    }
   ],
   "source": [
    "news_df[\"cat\"].value_counts()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "HcytXMG3h-y7"
   },
   "outputs": [],
   "source": [
    "# it looks like THE WORLDPOST and WORLDPOST are similar categories, so lets convert THE WORLDPOST to WORLDPOST\n",
    "news_df.loc[news_df[\"cat\"] == \"THE WORLDPOST\", \"cat\"] = \"WORLDPOST\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "hgXfIxsOlRcY"
   },
   "outputs": [],
   "source": [
    "# both 'ARTS & CULTURE' and 'CULTURE & ARTS' categories would represent the same kind of news, so lets convert them to a same tag.\n",
    "# also lets view the news with 'ARTS' tag and check whether they can be grouped into 'ARTS & CULTURE' category."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 141
    },
    "colab_type": "code",
    "id": "66OZccoHl6pg",
    "outputId": "ee78336d-eea3-42e9-9c74-f06e2ef7a762"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "6715    Modeling Agencies Enabled Sexual Predators For Years, Former Agent Says In the 1980s and '90s, Carolyn Kramer said she and her colleagues knew about allegedly predatory photographers, and sent the...\n",
       "7426    Actor Jeff Hiller Talks “Bright Colors And Bold Patterns” and More (AUDIO) This week I talked with actor Jeff Hiller about the hit Off Broadway play Bright Colors And Bold Patterns that he’ll be j...\n",
       "7698                                            New Yorker Cover Puts Trump 'In The Hole' After 'Racist' Comment The president reportedly referred to groups of immigrants as coming from \"shithole countries.\"\n",
       "7744                                                                                                              J. K. Rowling Trolls Trump For Canceled UK Visit With Clucking Great Emoji Not a scaredy-cat.\n",
       "7928                                                                                                     Man Surprises Girlfriend By Drawing Them In Different Animation Styles What a colorful Christmas gift.\n",
       "Name: text, dtype: object"
      ]
     },
     "execution_count": 24,
     "metadata": {
      "tags": []
     },
     "output_type": "execute_result"
    }
   ],
   "source": [
    "news_df[news_df['cat'] == 'ARTS & CULTURE']['text'].head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 141
    },
    "colab_type": "code",
    "id": "p5QM9C2PnXzL",
    "outputId": "9f9d8d11-82bd-4471-8791-df161ad2c73b"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "125440    First Nighter: Athol Fugard's Shadow..., Richard Maxwell's Isolde Although Athol Fugard turns 82 in June and his protagonist in The Shadow of the Hummingbird is already 84, the two-year difference...\n",
       "125441         Every Dance Move From A To Z Showcased In One Ridiculously Insane Video In this incredible video, a group of highly talented dancers demonstrates every dance style from A to Z on the rooftops of\n",
       "125442        'F*cking Tourist' Series Will Make You Think Twice About How You Travel Photographer Nicolas Demeersman makes the universal aversion to tourism very, very clear in his photography series \"Fucking\n",
       "125443                                  Stunning Photos Of A Homeless Boxer's Mansion See more photos on Slate. Even on a peninsula populated by handmade, ramshackle structures, Anderson’s place, half home and\n",
       "125518                          Raymond Pettibon's Surf Paintings Are Pure Punk Poetry In his current exhibition \"Are Your Motives Pure?\" Pettibon compiles surf-centric artworks from 1985 to 2013, ranging from\n",
       "Name: text, dtype: object"
      ]
     },
     "execution_count": 25,
     "metadata": {
      "tags": []
     },
     "output_type": "execute_result"
    }
   ],
   "source": [
    "news_df[news_df['cat'] == 'CULTURE & ARTS']['text'].head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 121
    },
    "colab_type": "code",
    "id": "q0U6VFnhnbnF",
    "outputId": "bbcebe48-663f-4d33-e582-5a34ad63d81a"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "36003    An Alert, Well-Hydrated Artist in No Acute Distress--Episode Thirty-Five: The Birth of a Book, the End of a Story \n",
       "37852                                                      Stage Door: Ute Lemper's Songs From The Broken Heart, Confucius \n",
       "38462                                                                                            Donna Quesada: Art Review \n",
       "38811                                                    Top 5 Sizzling Hot Winter Music Festivals in Frigid New York City \n",
       "38981                                                                          Defending their lives in 'Ride the Cyclone' \n",
       "Name: text, dtype: object"
      ]
     },
     "execution_count": 26,
     "metadata": {
      "tags": []
     },
     "output_type": "execute_result"
    }
   ],
   "source": [
    "news_df[news_df['cat'] == 'ARTS']['text'].head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "DlD560h5neK0"
   },
   "outputs": [],
   "source": [
    "# it seems like the news with 'ARTS' tag is only about entertainment industry news , whereas the 'ARTS & CULTURE'\n",
    "# news tell about arts in society. I am going to keep 'ART' as a seperate category."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "gDoj9okMnglK"
   },
   "outputs": [],
   "source": [
    "news_df.loc[news_df['cat'] == 'CULTURE & ARTS', 'cat'] = 'ARTS & CULTURE'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 293
    },
    "colab_type": "code",
    "id": "ZOjKZDVyni9n",
    "outputId": "cd108653-098d-458a-ee67-fd6597461997"
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1440x360 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light",
      "tags": []
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "news_count = news_df['cat'].value_counts()\n",
    "news_percent = [len(news_df[news_df['cat'] == news_cat]) / len(news_df) * 100 for news_cat in news_count.index]\n",
    "\n",
    "ax = news_count.plot(kind='bar', figsize=(20, 5))\n",
    "for idx, p in enumerate(ax.patches):\n",
    "    h = p.get_height()\n",
    "    x, y = p.get_xy()\n",
    "    ax.annotate('{:.1f}%'.format(news_percent[idx]), (x, y + h + 0.01))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "hzXn8DVnnpkd"
   },
   "outputs": [],
   "source": [
    "# from the above fig we can see that groups like 'EDUCATION', 'LATINO VOICES', 'COLLEGE' have very few examples \n",
    "# this may give us the problem of skewness or imbalanced classification. We will try to built a basic model to \n",
    "# check there is imbalanced classification problem and then use techniques like over sampling or under sampling \n",
    "# to tackle it."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "lplQJYJ2n-dH"
   },
   "outputs": [],
   "source": [
    "# now lets do some text preprocessing"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "c2Ld7ZT5oArH"
   },
   "outputs": [],
   "source": [
    "# lets do some text preprocessing\n",
    "REPLACE_BY_SPACE_RE = re.compile('[/(){}\\[\\]\\|@,;]')\n",
    "BAD_SYMBOLS_RE = re.compile('[^0-9a-z #+_]')\n",
    "STOPWORDS = set(stopwords.words('english'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "kMgtXdagog07"
   },
   "outputs": [],
   "source": [
    "def clean_text(text):\n",
    "    \n",
    "    text = text.lower()\n",
    "    text = REPLACE_BY_SPACE_RE.sub(' ', text) # replace REPLACE_BY_SPACE_RE symbols by space in text\n",
    "    text = BAD_SYMBOLS_RE.sub('', text) # remove BAD_SYMBOLS_RE symbols in text\n",
    "    text = ' '.join(word for word in text.split(' ') if word not in STOPWORDS) # remove stopwords from text\n",
    "    \n",
    "    return text"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "aXpzgKPPovy6"
   },
   "outputs": [],
   "source": [
    "news_df['text'] = news_df['text'].apply(clean_text)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "vG-csTqypGs5"
   },
   "outputs": [],
   "source": [
    "# tokenizing is the process of splitting the text into tokens. It create word_index for each token \n",
    "# which can be used to convert text sequence into integer sequence\n",
    "max_words = 10000\n",
    "tok = Tokenizer(num_words=max_words)\n",
    "tok.fit_on_texts(news_df['text'])\n",
    "sequence = tok.texts_to_sequences(news_df['text'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 34
    },
    "colab_type": "code",
    "id": "ODxufiZ1pULm",
    "outputId": "208a3e9f-f24f-4936-a038-be5bbc1146fc"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "200853"
      ]
     },
     "execution_count": 24,
     "metadata": {
      "tags": []
     },
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(sequence)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 52
    },
    "colab_type": "code",
    "id": "iZrrIKEnpXNu",
    "outputId": "ce1faa8c-0132-4654-94f1-2932be44d875"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "('2 mass shootings texas last week  1 tv left husband killed children another day america',\n",
       " 15)"
      ]
     },
     "execution_count": 38,
     "metadata": {
      "tags": []
     },
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# each of text in the news_df is converted to a sequence of words.\n",
    "news_df['text'][0], len(news_df['text'][0].split())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 34
    },
    "colab_type": "code",
    "id": "jf_f3QOFpjyw",
    "outputId": "08d5dea1-ce2a-4e67-ebca-28ffa04120fa"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "([211, 1179, 2777, 563, 52, 30, 186, 342, 365, 564, 449, 65, 95, 9, 111], 15)"
      ]
     },
     "execution_count": 39,
     "metadata": {
      "tags": []
     },
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sequence[0], len(sequence[0])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 34
    },
    "colab_type": "code",
    "id": "-lJpys3Kpn93",
    "outputId": "20d52bd3-0f46-4556-b257-9ad7e3781bac"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "('smith joins diplo nicky jam 2018 world cups official song course song', 12)"
      ]
     },
     "execution_count": 40,
     "metadata": {
      "tags": []
     },
     "output_type": "execute_result"
    }
   ],
   "source": [
    "news_df['text'][1], len(news_df['text'][1].split())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 34
    },
    "colab_type": "code",
    "id": "u2MaZxRzqp5v",
    "outputId": "a48ba08f-2d5a-4468-87c7-c3e763281a41"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "([1491, 2761, 5419, 2392, 15, 5878, 824, 835, 418, 835], 10)"
      ]
     },
     "execution_count": 41,
     "metadata": {
      "tags": []
     },
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sequence[1], len(sequence[1])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 34
    },
    "colab_type": "code",
    "id": "wwfkM_kLqsED",
    "outputId": "8d008dcc-99b2-4af1-e226-b3253a5b0e68"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Found 112533 unique tokens.\n"
     ]
    }
   ],
   "source": [
    "word_index = tok.word_index\n",
    "print(f'Found { len(word_index) } unique tokens.')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 258
    },
    "colab_type": "code",
    "id": "yoSsjM8aquhD",
    "outputId": "83bd5068-e6c5-40ec-fbaf-2dc80b0947e3"
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>text</th>\n",
       "      <th>cat</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>2 mass shootings texas last week  1 tv left husband killed children another day america</td>\n",
       "      <td>CRIME</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>smith joins diplo nicky jam 2018 world cups official song course song</td>\n",
       "      <td>ENTERTAINMENT</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>hugh grant marries first time age 57 actor longtime girlfriend anna eberstein tied knot civil ceremony</td>\n",
       "      <td>ENTERTAINMENT</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>jim carrey blasts castrato adam schiff democrats new artwork actor gives dems asskicking fighting hard enough donald trump</td>\n",
       "      <td>ENTERTAINMENT</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>julianna margulies uses donald trump poop bags pick dog dietland actress said using bags really cathartic  therapeutic moment</td>\n",
       "      <td>ENTERTAINMENT</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                                                                                                            text            cat\n",
       "0                                        2 mass shootings texas last week  1 tv left husband killed children another day america          CRIME\n",
       "1                                                          smith joins diplo nicky jam 2018 world cups official song course song  ENTERTAINMENT\n",
       "2                         hugh grant marries first time age 57 actor longtime girlfriend anna eberstein tied knot civil ceremony  ENTERTAINMENT\n",
       "3     jim carrey blasts castrato adam schiff democrats new artwork actor gives dems asskicking fighting hard enough donald trump  ENTERTAINMENT\n",
       "4  julianna margulies uses donald trump poop bags pick dog dietland actress said using bags really cathartic  therapeutic moment  ENTERTAINMENT"
      ]
     },
     "execution_count": 43,
     "metadata": {
      "tags": []
     },
     "output_type": "execute_result"
    }
   ],
   "source": [
    "news_df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "qtwXosr3qxDj"
   },
   "outputs": [],
   "source": [
    "# lets veiw the number of words in the articles on each new group\n",
    "news_df['word_seq'] = sequence\n",
    "news_df['num_words'] = news_df['word_seq'].str.len()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 1000
    },
    "colab_type": "code",
    "id": "TR8ISqFOq2qy",
    "outputId": "6a678c07-2b51-47c8-c01a-401336e2bfe0"
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>num_words</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>cat</th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>ARTS</th>\n",
       "      <td>14.363154</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>ARTS &amp; CULTURE</th>\n",
       "      <td>14.275222</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BLACK VOICES</th>\n",
       "      <td>14.802783</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BUSINESS</th>\n",
       "      <td>16.416035</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>COLLEGE</th>\n",
       "      <td>16.311189</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>COMEDY</th>\n",
       "      <td>12.389565</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CRIME</th>\n",
       "      <td>13.466960</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>DIVORCE</th>\n",
       "      <td>17.374781</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>EDUCATION</th>\n",
       "      <td>18.598606</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>ENTERTAINMENT</th>\n",
       "      <td>12.351227</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>ENVIRONMENT</th>\n",
       "      <td>18.223734</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>FIFTY</th>\n",
       "      <td>19.947181</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>FOOD &amp; DRINK</th>\n",
       "      <td>13.228236</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>GOOD NEWS</th>\n",
       "      <td>11.543634</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>GREEN</th>\n",
       "      <td>14.972921</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>HEALTHY LIVING</th>\n",
       "      <td>14.220795</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>HOME &amp; LIVING</th>\n",
       "      <td>15.483194</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>IMPACT</th>\n",
       "      <td>18.192252</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LATINO VOICES</th>\n",
       "      <td>14.085917</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>MEDIA</th>\n",
       "      <td>13.644405</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>MONEY</th>\n",
       "      <td>19.370826</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>PARENTING</th>\n",
       "      <td>18.779993</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>PARENTS</th>\n",
       "      <td>14.629836</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>POLITICS</th>\n",
       "      <td>14.900699</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>QUEER VOICES</th>\n",
       "      <td>15.627970</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>RELIGION</th>\n",
       "      <td>13.495305</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>SCIENCE</th>\n",
       "      <td>13.581726</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>SPORTS</th>\n",
       "      <td>12.772932</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>STYLE</th>\n",
       "      <td>9.828749</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>STYLE &amp; BEAUTY</th>\n",
       "      <td>17.280444</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TASTE</th>\n",
       "      <td>11.736641</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TECH</th>\n",
       "      <td>14.682517</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TRAVEL</th>\n",
       "      <td>17.244462</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>WEDDINGS</th>\n",
       "      <td>18.171734</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>WEIRD NEWS</th>\n",
       "      <td>9.799251</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>WELLNESS</th>\n",
       "      <td>19.095698</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>WOMEN</th>\n",
       "      <td>14.676504</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>WORLD NEWS</th>\n",
       "      <td>14.441893</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>WORLDPOST</th>\n",
       "      <td>14.502803</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                num_words\n",
       "cat                      \n",
       "ARTS            14.363154\n",
       "ARTS & CULTURE  14.275222\n",
       "BLACK VOICES    14.802783\n",
       "BUSINESS        16.416035\n",
       "COLLEGE         16.311189\n",
       "COMEDY          12.389565\n",
       "CRIME           13.466960\n",
       "DIVORCE         17.374781\n",
       "EDUCATION       18.598606\n",
       "ENTERTAINMENT   12.351227\n",
       "ENVIRONMENT     18.223734\n",
       "FIFTY           19.947181\n",
       "FOOD & DRINK    13.228236\n",
       "GOOD NEWS       11.543634\n",
       "GREEN           14.972921\n",
       "HEALTHY LIVING  14.220795\n",
       "HOME & LIVING   15.483194\n",
       "IMPACT          18.192252\n",
       "LATINO VOICES   14.085917\n",
       "MEDIA           13.644405\n",
       "MONEY           19.370826\n",
       "PARENTING       18.779993\n",
       "PARENTS         14.629836\n",
       "POLITICS        14.900699\n",
       "QUEER VOICES    15.627970\n",
       "RELIGION        13.495305\n",
       "SCIENCE         13.581726\n",
       "SPORTS          12.772932\n",
       "STYLE            9.828749\n",
       "STYLE & BEAUTY  17.280444\n",
       "TASTE           11.736641\n",
       "TECH            14.682517\n",
       "TRAVEL          17.244462\n",
       "WEDDINGS        18.171734\n",
       "WEIRD NEWS       9.799251\n",
       "WELLNESS        19.095698\n",
       "WOMEN           14.676504\n",
       "WORLD NEWS      14.441893\n",
       "WORLDPOST       14.502803"
      ]
     },
     "execution_count": 35,
     "metadata": {
      "tags": []
     },
     "output_type": "execute_result"
    }
   ],
   "source": [
    "news_df.groupby('cat').mean()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 729
    },
    "colab_type": "code",
    "id": "9pLrhw9Bq5Ey",
    "outputId": "f4e78b49-e15c-4c64-ec15-55cceb70b715"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "cat\n",
       "ARTS              1\n",
       "ARTS & CULTURE    1\n",
       "BLACK VOICES      0\n",
       "BUSINESS          0\n",
       "COLLEGE           0\n",
       "COMEDY            0\n",
       "CRIME             0\n",
       "DIVORCE           4\n",
       "EDUCATION         1\n",
       "ENTERTAINMENT     0\n",
       "ENVIRONMENT       5\n",
       "FIFTY             0\n",
       "FOOD & DRINK      2\n",
       "GOOD NEWS         0\n",
       "GREEN             0\n",
       "HEALTHY LIVING    0\n",
       "HOME & LIVING     2\n",
       "IMPACT            0\n",
       "LATINO VOICES     1\n",
       "MEDIA             0\n",
       "MONEY             8\n",
       "PARENTING         1\n",
       "PARENTS           1\n",
       "POLITICS          0\n",
       "QUEER VOICES      0\n",
       "RELIGION          0\n",
       "SCIENCE           0\n",
       "SPORTS            1\n",
       "STYLE             0\n",
       "STYLE & BEAUTY    1\n",
       "TASTE             0\n",
       "TECH              2\n",
       "TRAVEL            0\n",
       "WEDDINGS          5\n",
       "WEIRD NEWS        0\n",
       "WELLNESS          2\n",
       "WOMEN             0\n",
       "WORLD NEWS        3\n",
       "WORLDPOST         0\n",
       "Name: num_words, dtype: int64"
      ]
     },
     "execution_count": 46,
     "metadata": {
      "tags": []
     },
     "output_type": "execute_result"
    }
   ],
   "source": [
    "news_df.groupby('cat')['num_words'].min()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 493
    },
    "colab_type": "code",
    "id": "-08aPA1vq-oe",
    "outputId": "cc9b843d-ae4c-414e-a5c7-55d92d06ec0e"
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>text</th>\n",
       "      <th>cat</th>\n",
       "      <th>word_seq</th>\n",
       "      <th>num_words</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>557</th>\n",
       "      <td>yanny laurel heres real answer drumroll</td>\n",
       "      <td>WEIRD NEWS</td>\n",
       "      <td>[116, 101, 735]</td>\n",
       "      <td>3</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1486</th>\n",
       "      <td>watch paddleboarder get straightup wrecked dolphin dolphinately porpoise</td>\n",
       "      <td>WEIRD NEWS</td>\n",
       "      <td>[112, 10]</td>\n",
       "      <td>2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4871</th>\n",
       "      <td>pentagon devours budget normalizing budgetary bloat</td>\n",
       "      <td>POLITICS</td>\n",
       "      <td>[4712, 903]</td>\n",
       "      <td>2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6637</th>\n",
       "      <td>leaping kangaroo smashes unsuspecting cyclist australia kangaroo hopped away  apparently unscathed</td>\n",
       "      <td>WEIRD NEWS</td>\n",
       "      <td>[2461, 175, 1305]</td>\n",
       "      <td>3</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7142</th>\n",
       "      <td>doughnuteating champ arrested doughnut shop breakin doh</td>\n",
       "      <td>CRIME</td>\n",
       "      <td>[8677, 953, 1340]</td>\n",
       "      <td>3</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>191350</th>\n",
       "      <td>conversation designer bibhu mohapatra</td>\n",
       "      <td>STYLE &amp; BEAUTY</td>\n",
       "      <td>[746, 1057]</td>\n",
       "      <td>2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>191905</th>\n",
       "      <td>mothertomother real mothers</td>\n",
       "      <td>PARENTING</td>\n",
       "      <td>[101, 424]</td>\n",
       "      <td>2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>192309</th>\n",
       "      <td>defense london olympics</td>\n",
       "      <td>COMEDY</td>\n",
       "      <td>[1320, 750, 1217]</td>\n",
       "      <td>3</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>193389</th>\n",
       "      <td>brazilian decor</td>\n",
       "      <td>HOME &amp; LIVING</td>\n",
       "      <td>[5443, 3051]</td>\n",
       "      <td>2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>194036</th>\n",
       "      <td>video reclaimed design inspiration</td>\n",
       "      <td>HOME &amp; LIVING</td>\n",
       "      <td>[18, 1106, 1329]</td>\n",
       "      <td>3</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>3298 rows × 4 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "                                                                                                      text  ... num_words\n",
       "557                                                                yanny laurel heres real answer drumroll  ...         3\n",
       "1486                              watch paddleboarder get straightup wrecked dolphin dolphinately porpoise  ...         2\n",
       "4871                                                   pentagon devours budget normalizing budgetary bloat  ...         2\n",
       "6637    leaping kangaroo smashes unsuspecting cyclist australia kangaroo hopped away  apparently unscathed  ...         3\n",
       "7142                                               doughnuteating champ arrested doughnut shop breakin doh  ...         3\n",
       "...                                                                                                    ...  ...       ...\n",
       "191350                                                              conversation designer bibhu mohapatra   ...         2\n",
       "191905                                                                         mothertomother real mothers  ...         2\n",
       "192309                                                                            defense london olympics   ...         3\n",
       "193389                                                                                    brazilian decor   ...         2\n",
       "194036                                                                 video reclaimed design inspiration   ...         3\n",
       "\n",
       "[3298 rows x 4 columns]"
      ]
     },
     "execution_count": 64,
     "metadata": {
      "tags": []
     },
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# its seems like some articles does not have complete text\n",
    "news_df[news_df['num_words'] <= 3]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 34
    },
    "colab_type": "code",
    "id": "Hpdgd8D3rBQb",
    "outputId": "b42e01ac-d94e-42d7-9b18-42d582812bf6"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "3298"
      ]
     },
     "execution_count": 65,
     "metadata": {
      "tags": []
     },
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(news_df[news_df['num_words'] <= 3])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "pObcDpv9rDso"
   },
   "outputs": [],
   "source": [
    "# articles having few words are not providing sufficient information for classifying it into any group. \n",
    "# also the number of such samples is less, so we can just not consider them\n",
    "news_df = news_df[news_df['num_words'] > 3]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "PkGkrHvxrFwh"
   },
   "outputs": [],
   "source": [
    "x_train, x_test, y_train, y_test = train_test_split(news_df['text'], news_df['cat'], test_size=0.3, random_state=33)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "eKmozPBMrJaE"
   },
   "outputs": [],
   "source": [
    "# lets try our first baseline model using Naive Bayes Classifier"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 888
    },
    "colab_type": "code",
    "id": "kCPNZtVIrNf_",
    "outputId": "6b5b34ef-5919-4cbe-9c4f-7e9843e623ca"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "accuracy 0.3928830546509862 : \n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/usr/local/lib/python3.6/dist-packages/sklearn/metrics/_classification.py:1272: UndefinedMetricWarning: Precision and F-score are ill-defined and being set to 0.0 in labels with no predicted samples. Use `zero_division` parameter to control this behavior.\n",
      "  _warn_prf(average, modifier, msg_start, len(result))\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "                precision    recall  f1-score   support\n",
      "\n",
      "          ARTS       0.00      0.00      0.00       408\n",
      "ARTS & CULTURE       1.00      0.00      0.00       681\n",
      "  BLACK VOICES       0.00      0.00      0.00      1261\n",
      "      BUSINESS       0.85      0.02      0.03      1728\n",
      "       COLLEGE       0.00      0.00      0.00       367\n",
      "        COMEDY       1.00      0.00      0.01      1436\n",
      "         CRIME       0.76      0.01      0.03       987\n",
      "       DIVORCE       1.00      0.04      0.08      1049\n",
      "     EDUCATION       0.00      0.00      0.00       311\n",
      " ENTERTAINMENT       0.50      0.72      0.59      4797\n",
      "   ENVIRONMENT       0.00      0.00      0.00       425\n",
      "         FIFTY       0.00      0.00      0.00       406\n",
      "  FOOD & DRINK       0.72      0.41      0.53      1900\n",
      "     GOOD NEWS       0.00      0.00      0.00       394\n",
      "         GREEN       0.00      0.00      0.00       750\n",
      "HEALTHY LIVING       1.00      0.00      0.00      1913\n",
      " HOME & LIVING       0.96      0.13      0.23      1255\n",
      "        IMPACT       0.00      0.00      0.00      1030\n",
      " LATINO VOICES       0.00      0.00      0.00       337\n",
      "         MEDIA       0.00      0.00      0.00       860\n",
      "         MONEY       0.00      0.00      0.00       533\n",
      "     PARENTING       0.54      0.18      0.27      2593\n",
      "       PARENTS       1.00      0.01      0.02      1179\n",
      "      POLITICS       0.31      0.99      0.47      9631\n",
      "  QUEER VOICES       0.94      0.07      0.13      1925\n",
      "      RELIGION       1.00      0.00      0.00       720\n",
      "       SCIENCE       1.00      0.01      0.01       628\n",
      "        SPORTS       0.94      0.09      0.17      1413\n",
      "         STYLE       0.00      0.00      0.00       660\n",
      "STYLE & BEAUTY       0.69      0.68      0.68      2848\n",
      "         TASTE       0.00      0.00      0.00       601\n",
      "          TECH       1.00      0.01      0.02       622\n",
      "        TRAVEL       0.74      0.55      0.63      2936\n",
      "      WEDDINGS       0.94      0.13      0.23      1136\n",
      "    WEIRD NEWS       0.00      0.00      0.00       788\n",
      "      WELLNESS       0.33      0.88      0.48      5282\n",
      "         WOMEN       1.00      0.03      0.05      1011\n",
      "    WORLD NEWS       0.00      0.00      0.00       635\n",
      "     WORLDPOST       0.83      0.07      0.13      1831\n",
      "\n",
      "      accuracy                           0.39     59267\n",
      "     macro avg       0.49      0.13      0.12     59267\n",
      "  weighted avg       0.55      0.39      0.29     59267\n",
      "\n"
     ]
    }
   ],
   "source": [
    "nb = Pipeline([\n",
    "    ('vect', CountVectorizer()),\n",
    "    ('tfidf', TfidfTransformer()),\n",
    "    ('clf', MultinomialNB())\n",
    "])\n",
    "\n",
    "nb.fit(x_train, y_train)\n",
    "y_pred = nb.predict(x_test)\n",
    "\n",
    "print('accuracy %s : '% accuracy_score(y_pred, y_test))\n",
    "print(classification_report(y_test, y_pred))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "nfwPKu5TrZO8"
   },
   "outputs": [],
   "source": [
    "# Naive_bayes classification gives us an accuracy of 39%. With this as base benchmark lets try improving accuracy\n",
    "# using other models"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "TE4Dscbsrgjo"
   },
   "outputs": [],
   "source": [
    "# Linear Support Vector Machine"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 833
    },
    "colab_type": "code",
    "id": "OVY0LlNVriqG",
    "outputId": "c7915f6a-d25c-4f13-aaea-84331e974349"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "accuracy 0.5595862790423001 : \n",
      "                precision    recall  f1-score   support\n",
      "\n",
      "          ARTS       0.28      0.09      0.14       408\n",
      "ARTS & CULTURE       0.41      0.23      0.30       681\n",
      "  BLACK VOICES       0.49      0.25      0.33      1261\n",
      "      BUSINESS       0.50      0.33      0.40      1728\n",
      "       COLLEGE       0.45      0.23      0.31       367\n",
      "        COMEDY       0.47      0.19      0.28      1436\n",
      "         CRIME       0.54      0.43      0.48       987\n",
      "       DIVORCE       0.74      0.74      0.74      1049\n",
      "     EDUCATION       0.50      0.20      0.29       311\n",
      " ENTERTAINMENT       0.60      0.69      0.64      4797\n",
      "   ENVIRONMENT       0.44      0.15      0.22       425\n",
      "         FIFTY       0.36      0.14      0.20       406\n",
      "  FOOD & DRINK       0.57      0.72      0.64      1900\n",
      "     GOOD NEWS       0.30      0.12      0.17       394\n",
      "         GREEN       0.41      0.19      0.26       750\n",
      "HEALTHY LIVING       0.36      0.06      0.11      1913\n",
      " HOME & LIVING       0.64      0.63      0.63      1255\n",
      "        IMPACT       0.45      0.17      0.25      1030\n",
      " LATINO VOICES       0.59      0.19      0.28       337\n",
      "         MEDIA       0.56      0.24      0.33       860\n",
      "         MONEY       0.53      0.23      0.32       533\n",
      "     PARENTING       0.52      0.53      0.53      2593\n",
      "       PARENTS       0.38      0.08      0.13      1179\n",
      "      POLITICS       0.54      0.92      0.68      9631\n",
      "  QUEER VOICES       0.71      0.67      0.69      1925\n",
      "      RELIGION       0.57      0.37      0.45       720\n",
      "       SCIENCE       0.60      0.37      0.45       628\n",
      "        SPORTS       0.68      0.65      0.66      1413\n",
      "         STYLE       0.30      0.04      0.07       660\n",
      "STYLE & BEAUTY       0.56      0.81      0.66      2848\n",
      "         TASTE       0.18      0.03      0.06       601\n",
      "          TECH       0.57      0.33      0.42       622\n",
      "        TRAVEL       0.62      0.77      0.69      2936\n",
      "      WEDDINGS       0.72      0.79      0.75      1136\n",
      "    WEIRD NEWS       0.42      0.15      0.22       788\n",
      "      WELLNESS       0.52      0.80      0.63      5282\n",
      "         WOMEN       0.40      0.18      0.25      1011\n",
      "    WORLD NEWS       0.46      0.08      0.13       635\n",
      "     WORLDPOST       0.59      0.39      0.47      1831\n",
      "\n",
      "      accuracy                           0.56     59267\n",
      "     macro avg       0.50      0.36      0.39     59267\n",
      "  weighted avg       0.54      0.56      0.51     59267\n",
      "\n"
     ]
    }
   ],
   "source": [
    "sgd = Pipeline([\n",
    "    ('vect', CountVectorizer()),\n",
    "    ('tfidf', TfidfTransformer()),\n",
    "    ('clf', SGDClassifier(loss='hinge', penalty='l2', alpha=1e-3, random_state=42, max_iter=5, tol=None))\n",
    "])\n",
    "\n",
    "sgd.fit(x_train, y_train)\n",
    "y_pred = sgd.predict(x_test)\n",
    "\n",
    "print('accuracy %s : ' % accuracy_score(y_pred, y_test))\n",
    "print(classification_report(y_test, y_pred))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "ZWmbillJrk-N"
   },
   "outputs": [],
   "source": [
    "# deep learning model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 173
    },
    "colab_type": "code",
    "id": "93feMQUxxaAV",
    "outputId": "98964e41-fb0b-4d3c-818c-1a15b6cb2857"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "count    197555.000000\n",
       "mean         15.549862\n",
       "std           6.558696\n",
       "min           4.000000\n",
       "25%          11.000000\n",
       "50%          15.000000\n",
       "75%          19.000000\n",
       "max         127.000000\n",
       "Name: num_words, dtype: float64"
      ]
     },
     "execution_count": 50,
     "metadata": {
      "tags": []
     },
     "output_type": "execute_result"
    }
   ],
   "source": [
    "news_df['num_words'].describe()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "JBApQZj_xc4X"
   },
   "outputs": [],
   "source": [
    "# our model expects the lenght of all sequence to be same. Now the lenght of the sequences are different corresponding\n",
    "# to the length of the actual text. We can generalize it to a specific length of 60 by padding.\n",
    "max_len = 30\n",
    "data = pad_sequences(news_df['word_seq'], maxlen=max_len)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 69
    },
    "colab_type": "code",
    "id": "BhWPd39VxfaB",
    "outputId": "92731f9c-9057-4db7-dfa7-d6f9da99d5e0"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([   0,    0,    0,    0,    0,    0,    0,    0,    0,    0,    0,\n",
       "          0,    0,    0,    0,  211, 1179, 2777,  563,   52,   30,  186,\n",
       "        342,  365,  564,  449,   65,   95,    9,  111], dtype=int32)"
      ]
     },
     "execution_count": 28,
     "metadata": {
      "tags": []
     },
     "output_type": "execute_result"
    }
   ],
   "source": [
    "data[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "HvPjVCdKxiGG"
   },
   "outputs": [],
   "source": [
    "# one hot encode the labels\n",
    "encoder = LabelEncoder()\n",
    "encoder.fit(news_df['cat'])\n",
    "labels = encoder.transform(news_df['cat'])\n",
    "n_classes = max(labels) + 1\n",
    "labels = to_categorical(labels, n_classes)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 34
    },
    "colab_type": "code",
    "id": "rp5yD6cLxrlg",
    "outputId": "569656ea-e878-4454-b3f8-cf9e8a02a58e"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(197555, 39)"
      ]
     },
     "execution_count": 30,
     "metadata": {
      "tags": []
     },
     "output_type": "execute_result"
    }
   ],
   "source": [
    "labels.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "UxOpjalRxuHu"
   },
   "outputs": [],
   "source": [
    "# split the data into train, validation and test sets"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "6TqQTD1zxwr2"
   },
   "outputs": [],
   "source": [
    "x_train, x_test, y_train, y_test = train_test_split(data, labels, test_size=0.3, random_state=20)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 34
    },
    "colab_type": "code",
    "id": "dpmpE-5ox0FP",
    "outputId": "72152d64-9a20-48a8-e513-acb8ad768ce1"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((138288, 30), (59267, 30))"
      ]
     },
     "execution_count": 33,
     "metadata": {
      "tags": []
     },
     "output_type": "execute_result"
    }
   ],
   "source": [
    "x_train.shape, x_test.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "RmnavV9bx2r-"
   },
   "outputs": [],
   "source": [
    "x_train, x_val, y_train, y_val = train_test_split(x_train, y_train, test_size=0.2, random_state=20)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 34
    },
    "colab_type": "code",
    "id": "_xF9MfrWx51J",
    "outputId": "2bfa524e-f216-4a5e-97a2-30075f269daa"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((110630, 30), (27658, 30))"
      ]
     },
     "execution_count": 35,
     "metadata": {
      "tags": []
     },
     "output_type": "execute_result"
    }
   ],
   "source": [
    "x_train.shape, x_val.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "y4i821OYx8qx"
   },
   "outputs": [],
   "source": [
    "# building a basic and cheap model\n",
    "\n",
    "model = Sequential()\n",
    "model.add(Embedding(max_words, 64, input_length=max_len))\n",
    "model.add(Flatten())\n",
    "model.add(Dense(64, activation='relu'))\n",
    "model.add(Dense(n_classes, activation='softmax'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 295
    },
    "colab_type": "code",
    "id": "eKEcvLNFyDNa",
    "outputId": "9a44d151-31d4-4697-a874-0eb0b12541cb"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential_4\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "embedding_4 (Embedding)      (None, 30, 64)            640000    \n",
      "_________________________________________________________________\n",
      "flatten_4 (Flatten)          (None, 1920)              0         \n",
      "_________________________________________________________________\n",
      "dense_7 (Dense)              (None, 64)                122944    \n",
      "_________________________________________________________________\n",
      "dense_8 (Dense)              (None, 39)                2535      \n",
      "=================================================================\n",
      "Total params: 765,479\n",
      "Trainable params: 765,479\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "-xcdAn1-yFxo"
   },
   "outputs": [],
   "source": [
    "model.compile(optimizer='rmsprop', loss='categorical_crossentropy', metrics=['accuracy'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 263
    },
    "colab_type": "code",
    "id": "GIJeF-_JzDz7",
    "outputId": "b61f9a55-e03f-4eac-830c-5cf439a1036d"
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/usr/local/lib/python3.6/dist-packages/tensorflow/python/framework/indexed_slices.py:434: UserWarning: Converting sparse IndexedSlices to a dense Tensor of unknown shape. This may consume a large amount of memory.\n",
      "  \"Converting sparse IndexedSlices to a dense Tensor of unknown shape. \"\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 110630 samples, validate on 27658 samples\n",
      "Epoch 1/5\n",
      "110630/110630 [==============================] - 9s 80us/step - loss: 2.3897 - accuracy: 0.3894 - val_loss: 1.8844 - val_accuracy: 0.5071\n",
      "Epoch 2/5\n",
      "110630/110630 [==============================] - 9s 79us/step - loss: 1.6066 - accuracy: 0.5709 - val_loss: 1.6172 - val_accuracy: 0.5676\n",
      "Epoch 3/5\n",
      "110630/110630 [==============================] - 9s 79us/step - loss: 1.3224 - accuracy: 0.6367 - val_loss: 1.5535 - val_accuracy: 0.5832\n",
      "Epoch 4/5\n",
      "110630/110630 [==============================] - 9s 77us/step - loss: 1.1025 - accuracy: 0.6941 - val_loss: 1.5717 - val_accuracy: 0.5820\n",
      "Epoch 5/5\n",
      "110630/110630 [==============================] - 9s 77us/step - loss: 0.8985 - accuracy: 0.7521 - val_loss: 1.6513 - val_accuracy: 0.5760\n"
     ]
    }
   ],
   "source": [
    "history = model.fit(\n",
    "                x_train, \n",
    "                y_train,\n",
    "                epochs=5,\n",
    "                batch_size=128,\n",
    "                validation_data=(x_val, y_val)\n",
    "            )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "SGB1gh1CzGfc"
   },
   "outputs": [],
   "source": [
    "def plot_model(history, epch):\n",
    "  acc = history.history['accuracy']\n",
    "  val_acc = history.history['val_accuracy']\n",
    "  loss = history.history['loss']\n",
    "  val_loss = history.history['val_loss']\n",
    "\n",
    "  epochs = range(1, epch+1)\n",
    "\n",
    "  plt.plot(epochs, acc, 'bo', label='Training_acc')\n",
    "  plt.plot(epochs, val_acc, 'b', label='Validation_acc')\n",
    "  plt.title('Training and validation accuracy')\n",
    "  plt.legend()\n",
    "\n",
    "  plt.figure()\n",
    "  plt.plot(epochs, loss, 'bo', label='Training_loss')\n",
    "  plt.plot(epochs, val_loss, 'b', label='Validation_loss')\n",
    "  plt.title('Training and Validation loss')\n",
    "  plt.legend()\n",
    "\n",
    "  plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 545
    },
    "colab_type": "code",
    "id": "l6ixkTbl0KWU",
    "outputId": "d824b246-3c47-4e5a-f873-6dfa6b1c8cb9"
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light",
      "tags": []
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light",
      "tags": []
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plot_model(history, 5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "GMBXn9Cx0QMD"
   },
   "outputs": [],
   "source": [
    "# our model starts to overfit after some epochs. Lets try introducing some regularization."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 52
    },
    "colab_type": "code",
    "id": "QfU0BcAk0cbL",
    "outputId": "d58e3f86-bcd4-4d76-b86e-940e8883c8a4"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "59267/59267 [==============================] - 2s 35us/step\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[1.6675115256069992, 0.5723589658737183]"
      ]
     },
     "execution_count": 113,
     "metadata": {
      "tags": []
     },
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.evaluate(x_test, y_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "66mabqUG0eym"
   },
   "outputs": [],
   "source": [
    "# we will try introducing regularization to handle overfitting\n",
    "\n",
    "model = Sequential()\n",
    "model.add(Embedding(max_words, 64, input_length=max_len))\n",
    "model.add(Flatten())\n",
    "model.add(Dense(64, activation='relu', kernel_regularizer=regularizers.l2(0.01)))\n",
    "model.add(Dense(n_classes, activation='softmax'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 295
    },
    "colab_type": "code",
    "id": "Zz5wN1kc0opK",
    "outputId": "49a94455-3ad9-4290-c3dd-a5e261c4d277"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential_6\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "embedding_6 (Embedding)      (None, 30, 64)            640000    \n",
      "_________________________________________________________________\n",
      "flatten_6 (Flatten)          (None, 1920)              0         \n",
      "_________________________________________________________________\n",
      "dense_11 (Dense)             (None, 64)                122944    \n",
      "_________________________________________________________________\n",
      "dense_12 (Dense)             (None, 39)                2535      \n",
      "=================================================================\n",
      "Total params: 765,479\n",
      "Trainable params: 765,479\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "bO7qLRDA0qlr"
   },
   "outputs": [],
   "source": [
    "model.compile(optimizer='rmsprop', loss='categorical_crossentropy', metrics=['accuracy'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 627
    },
    "colab_type": "code",
    "id": "i__fQyh90sq6",
    "outputId": "365ce4c5-d8d4-48e3-e8c1-1db3db90405a"
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/usr/local/lib/python3.6/dist-packages/tensorflow/python/framework/indexed_slices.py:434: UserWarning: Converting sparse IndexedSlices to a dense Tensor of unknown shape. This may consume a large amount of memory.\n",
      "  \"Converting sparse IndexedSlices to a dense Tensor of unknown shape. \"\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 110630 samples, validate on 27658 samples\n",
      "Epoch 1/15\n",
      "110630/110630 [==============================] - 9s 81us/step - loss: 2.8677 - accuracy: 0.2962 - val_loss: 2.5818 - val_accuracy: 0.3464\n",
      "Epoch 2/15\n",
      "110630/110630 [==============================] - 9s 79us/step - loss: 2.4183 - accuracy: 0.3852 - val_loss: 2.3378 - val_accuracy: 0.4114\n",
      "Epoch 3/15\n",
      "110630/110630 [==============================] - 9s 80us/step - loss: 2.1689 - accuracy: 0.4576 - val_loss: 2.1303 - val_accuracy: 0.4772\n",
      "Epoch 4/15\n",
      "110630/110630 [==============================] - 9s 78us/step - loss: 1.9760 - accuracy: 0.5183 - val_loss: 1.9927 - val_accuracy: 0.5205\n",
      "Epoch 5/15\n",
      "110630/110630 [==============================] - 9s 78us/step - loss: 1.8332 - accuracy: 0.5505 - val_loss: 1.9095 - val_accuracy: 0.5329\n",
      "Epoch 6/15\n",
      "110630/110630 [==============================] - 9s 78us/step - loss: 1.7344 - accuracy: 0.5729 - val_loss: 1.8486 - val_accuracy: 0.5414\n",
      "Epoch 7/15\n",
      "110630/110630 [==============================] - 9s 80us/step - loss: 1.6592 - accuracy: 0.5913 - val_loss: 1.8330 - val_accuracy: 0.5478\n",
      "Epoch 8/15\n",
      "110630/110630 [==============================] - 9s 79us/step - loss: 1.5969 - accuracy: 0.6056 - val_loss: 1.8226 - val_accuracy: 0.5571\n",
      "Epoch 9/15\n",
      "110630/110630 [==============================] - 9s 78us/step - loss: 1.5373 - accuracy: 0.6230 - val_loss: 1.8265 - val_accuracy: 0.5566\n",
      "Epoch 10/15\n",
      "110630/110630 [==============================] - 8s 76us/step - loss: 1.4803 - accuracy: 0.6405 - val_loss: 1.8218 - val_accuracy: 0.5603\n",
      "Epoch 11/15\n",
      "110630/110630 [==============================] - 8s 76us/step - loss: 1.4229 - accuracy: 0.6588 - val_loss: 1.8727 - val_accuracy: 0.5535\n",
      "Epoch 12/15\n",
      "110630/110630 [==============================] - 8s 76us/step - loss: 1.3640 - accuracy: 0.6776 - val_loss: 1.9057 - val_accuracy: 0.5479\n",
      "Epoch 13/15\n",
      "110630/110630 [==============================] - 8s 76us/step - loss: 1.3046 - accuracy: 0.6975 - val_loss: 1.9213 - val_accuracy: 0.5486\n",
      "Epoch 14/15\n",
      "110630/110630 [==============================] - 8s 76us/step - loss: 1.2450 - accuracy: 0.7170 - val_loss: 1.9766 - val_accuracy: 0.5490\n",
      "Epoch 15/15\n",
      "110630/110630 [==============================] - 8s 76us/step - loss: 1.1875 - accuracy: 0.7347 - val_loss: 2.0454 - val_accuracy: 0.5336\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<keras.callbacks.callbacks.History at 0x7fc67f911f28>"
      ]
     },
     "execution_count": 121,
     "metadata": {
      "tags": []
     },
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.fit(\n",
    "    x_train,\n",
    "    y_train,\n",
    "    epochs=15,\n",
    "    batch_size=128,\n",
    "    validation_data=(x_val, y_val)\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "iFb0yxZL0ya5"
   },
   "outputs": [],
   "source": [
    "# it looks like our model is not able to get accuracy above 56%, lets try adding more layers and see the performance."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "yjxw7ZvU4QxW"
   },
   "outputs": [],
   "source": [
    "model = Sequential()\n",
    "model.add(Embedding(max_words, 64, input_length=max_len))\n",
    "model.add(Flatten())\n",
    "model.add(Dense(64, activation=\"relu\", kernel_regularizer=regularizers.l2(0.01)))\n",
    "model.add(Dense(64, activation=\"relu\", kernel_regularizer=regularizers.l2(0.01)))\n",
    "model.add(Dense(n_classes, activation=\"softmax\"))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 330
    },
    "colab_type": "code",
    "id": "SUKVslLy43VC",
    "outputId": "0a14004b-d917-4483-d85a-99c2134423a2"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential_11\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "embedding_10 (Embedding)     (None, 30, 64)            640000    \n",
      "_________________________________________________________________\n",
      "flatten_10 (Flatten)         (None, 1920)              0         \n",
      "_________________________________________________________________\n",
      "dense_19 (Dense)             (None, 64)                122944    \n",
      "_________________________________________________________________\n",
      "dense_20 (Dense)             (None, 64)                4160      \n",
      "_________________________________________________________________\n",
      "dense_21 (Dense)             (None, 39)                2535      \n",
      "=================================================================\n",
      "Total params: 769,639\n",
      "Trainable params: 769,639\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "UKyvDM4W5B_S"
   },
   "outputs": [],
   "source": [
    "model.compile(\n",
    "    optimizer=\"rmsprop\", \n",
    "    loss=\"categorical_crossentropy\",\n",
    "    metrics=[\"accuracy\"]\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 801
    },
    "colab_type": "code",
    "id": "kBLJeiKj5TgF",
    "outputId": "e319c348-0b83-4437-d8d6-d6054e705b64"
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/usr/local/lib/python3.6/dist-packages/tensorflow/python/framework/indexed_slices.py:434: UserWarning: Converting sparse IndexedSlices to a dense Tensor of unknown shape. This may consume a large amount of memory.\n",
      "  \"Converting sparse IndexedSlices to a dense Tensor of unknown shape. \"\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 110630 samples, validate on 27658 samples\n",
      "Epoch 1/20\n",
      "110630/110630 [==============================] - 9s 80us/step - loss: 2.9860 - accuracy: 0.2731 - val_loss: 2.6970 - val_accuracy: 0.3297\n",
      "Epoch 2/20\n",
      "110630/110630 [==============================] - 9s 78us/step - loss: 2.5576 - accuracy: 0.3512 - val_loss: 2.5391 - val_accuracy: 0.3531\n",
      "Epoch 3/20\n",
      "110630/110630 [==============================] - 9s 82us/step - loss: 2.4301 - accuracy: 0.3724 - val_loss: 2.4511 - val_accuracy: 0.3741\n",
      "Epoch 4/20\n",
      "110630/110630 [==============================] - 9s 78us/step - loss: 2.3146 - accuracy: 0.4122 - val_loss: 2.3369 - val_accuracy: 0.4168\n",
      "Epoch 5/20\n",
      "110630/110630 [==============================] - 9s 78us/step - loss: 2.1937 - accuracy: 0.4461 - val_loss: 2.2803 - val_accuracy: 0.4293\n",
      "Epoch 6/20\n",
      "110630/110630 [==============================] - 9s 78us/step - loss: 2.1120 - accuracy: 0.4674 - val_loss: 2.2430 - val_accuracy: 0.4482\n",
      "Epoch 7/20\n",
      "110630/110630 [==============================] - 9s 77us/step - loss: 2.0391 - accuracy: 0.4891 - val_loss: 2.2553 - val_accuracy: 0.4495\n",
      "Epoch 8/20\n",
      "110630/110630 [==============================] - 9s 77us/step - loss: 1.9623 - accuracy: 0.5124 - val_loss: 2.2598 - val_accuracy: 0.4611\n",
      "Epoch 9/20\n",
      "110630/110630 [==============================] - 9s 77us/step - loss: 1.8833 - accuracy: 0.5358 - val_loss: 2.2803 - val_accuracy: 0.4617\n",
      "Epoch 10/20\n",
      "110630/110630 [==============================] - 8s 77us/step - loss: 1.8036 - accuracy: 0.5583 - val_loss: 2.3495 - val_accuracy: 0.4600\n",
      "Epoch 11/20\n",
      "110630/110630 [==============================] - 8s 76us/step - loss: 1.7183 - accuracy: 0.5855 - val_loss: 2.3854 - val_accuracy: 0.4660\n",
      "Epoch 12/20\n",
      "110630/110630 [==============================] - 8s 77us/step - loss: 1.6290 - accuracy: 0.6127 - val_loss: 2.4632 - val_accuracy: 0.4576\n",
      "Epoch 13/20\n",
      "110630/110630 [==============================] - 9s 78us/step - loss: 1.5395 - accuracy: 0.6404 - val_loss: 2.5792 - val_accuracy: 0.4462\n",
      "Epoch 14/20\n",
      "110630/110630 [==============================] - 9s 77us/step - loss: 1.4528 - accuracy: 0.6694 - val_loss: 2.6761 - val_accuracy: 0.4368\n",
      "Epoch 15/20\n",
      "110630/110630 [==============================] - 8s 77us/step - loss: 1.3732 - accuracy: 0.6961 - val_loss: 2.7554 - val_accuracy: 0.4427\n",
      "Epoch 16/20\n",
      "110630/110630 [==============================] - 9s 77us/step - loss: 1.2951 - accuracy: 0.7214 - val_loss: 2.8782 - val_accuracy: 0.4354\n",
      "Epoch 17/20\n",
      "110630/110630 [==============================] - 9s 77us/step - loss: 1.2254 - accuracy: 0.7434 - val_loss: 2.9839 - val_accuracy: 0.4315\n",
      "Epoch 18/20\n",
      "110630/110630 [==============================] - 9s 77us/step - loss: 1.1596 - accuracy: 0.7642 - val_loss: 3.1225 - val_accuracy: 0.4262\n",
      "Epoch 19/20\n",
      "110630/110630 [==============================] - 9s 78us/step - loss: 1.1012 - accuracy: 0.7827 - val_loss: 3.2634 - val_accuracy: 0.4153\n",
      "Epoch 20/20\n",
      "110630/110630 [==============================] - 9s 79us/step - loss: 1.0463 - accuracy: 0.7981 - val_loss: 3.3917 - val_accuracy: 0.4123\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<keras.callbacks.callbacks.History at 0x7fc67f4bb9b0>"
      ]
     },
     "execution_count": 137,
     "metadata": {
      "tags": []
     },
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.fit(\n",
    "    x_train, \n",
    "    y_train,\n",
    "    epochs=20,\n",
    "    batch_size=128,\n",
    "    validation_data=(x_val, y_val)\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 104
    },
    "colab_type": "code",
    "id": "0gSrnEXq5her",
    "outputId": "62c6a682-3362-4eb8-9ce3-011664a9952c"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "  % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current\n",
      "                                 Dload  Upload   Total   Spent    Left  Speed\n",
      "\r",
      "  0     0    0     0    0     0      0      0 --:--:-- --:--:-- --:--:--     0\r",
      "  0   308    0     0    0     0      0      0 --:--:-- --:--:-- --:--:--     0\n",
      "  0   345    0     0    0     0      0      0 --:--:-- --:--:-- --:--:--     0\n",
      "100  822M  100  822M    0     0  2165k      0  0:06:28  0:06:28 --:--:-- 2176k\n"
     ]
    }
   ],
   "source": [
    "# lets now try using pretrained word embedding - GloVe word embedding\n",
    "!curl -LO http://nlp.stanford.edu/data/glove.6B.zip"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 104
    },
    "colab_type": "code",
    "id": "kUemrM6m89yy",
    "outputId": "d4e2e7ae-a77c-4313-a2c1-f264b983ebfe"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Archive:  glove.6B.zip\n",
      "  inflating: glove.6B.50d.txt        \n",
      "  inflating: glove.6B.100d.txt       \n",
      "  inflating: glove.6B.200d.txt       \n",
      "  inflating: glove.6B.300d.txt       \n"
     ]
    }
   ],
   "source": [
    "!unzip glove.6B.zip"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "3pMZiqAX9FdL"
   },
   "outputs": [],
   "source": [
    "embeddings_index = {}\n",
    "\n",
    "with open(\"glove.6B.100d.txt\", \"r\") as file:\n",
    "    for line in file:\n",
    "        values = line.split()\n",
    "        word = values[0]\n",
    "        coeffs = np.asarray(values[1:], dtype='float32')\n",
    "        embeddings_index[word] = coeffs"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 34
    },
    "colab_type": "code",
    "id": "9587ttyB8CPh",
    "outputId": "17a8f249-75c2-4674-aa6d-37fb0aa0614a"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "there are 400000 tokens.\n"
     ]
    }
   ],
   "source": [
    "print(f\"there are {len(embeddings_index)} tokens.\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "ewcR2HWWJnk1"
   },
   "outputs": [],
   "source": [
    "# creating the embedding matrix\n",
    "embedding_dim = 100\n",
    "\n",
    "embedding_matrix = np.zeros((max_words, embedding_dim))\n",
    "for word, i in word_index.items():\n",
    "    if i < max_words:\n",
    "        embedding_vector = embeddings_index.get(word)\n",
    "        if embedding_vector is not None:\n",
    "            embedding_matrix[i] = embedding_vector"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 364
    },
    "colab_type": "code",
    "id": "CgO5l1SAJglg",
    "outputId": "0e99584f-decf-42ec-ea32-a91143970d8e"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([-0.22556999,  0.49417999,  0.48609999, -0.4332    ,  0.13738   ,\n",
       "        0.50616997,  0.26058   ,  0.30103001, -0.091486  ,  0.10876   ,\n",
       "        0.30579999,  0.051028  ,  0.22303   ,  0.054236  ,  0.068838  ,\n",
       "       -0.24700999,  0.32688999, -0.082203  , -0.28865999,  0.3734    ,\n",
       "        0.73803997, -0.040969  ,  0.040201  ,  0.11384   ,  0.69986999,\n",
       "       -0.49744999, -0.06755   , -0.42598999, -0.10725   , -0.010697  ,\n",
       "       -0.01479   ,  0.55975997,  0.3064    ,  0.053053  ,  0.058034  ,\n",
       "        0.32756001, -0.37233001,  0.46513   ,  0.14285   , -0.085003  ,\n",
       "       -0.45475999,  0.19773   ,  0.6383    , -0.31147999,  0.10858   ,\n",
       "        0.31557   ,  0.36682001, -0.35135001, -0.48414001, -0.33234999,\n",
       "       -0.33816001, -0.39678001,  0.1908    ,  1.3513    , -0.39043999,\n",
       "       -2.87949991, -0.14275999, -0.087754  ,  1.77129996,  0.99331999,\n",
       "       -0.14128999,  0.94388998,  0.050897  ,  0.47373   ,  0.86387002,\n",
       "       -0.16162001,  0.67198998,  0.52344   ,  0.14438   , -0.055194  ,\n",
       "       -0.34669   , -0.20742001,  0.18907   , -0.19845   ,  0.34862   ,\n",
       "        0.10121   , -0.092119  , -0.66258001, -1.0582    , -0.11803   ,\n",
       "        0.70170999,  0.077776  , -0.50546002,  0.032243  , -1.61759996,\n",
       "       -0.29302001, -0.061748  , -0.32473001,  0.3439    , -0.44698   ,\n",
       "        0.085689  ,  0.13294999, -0.1807    , -0.11854   , -0.82985002,\n",
       "        0.13784   , -0.34358999, -0.45743999,  0.49645999,  0.34906   ])"
      ]
     },
     "execution_count": 153,
     "metadata": {
      "tags": []
     },
     "output_type": "execute_result"
    }
   ],
   "source": [
    "embedding_matrix[2]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "lIFcUjnMKFUX"
   },
   "outputs": [],
   "source": [
    "model = Sequential()\n",
    "model.add(Embedding(max_words, embedding_dim, input_length=max_len))\n",
    "model.add(Flatten())\n",
    "model.add(Dense(64, activation=\"relu\"))\n",
    "model.add(Dense(n_classes, activation=\"softmax\"))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 295
    },
    "colab_type": "code",
    "id": "QV5GoD1nKOUE",
    "outputId": "0d070ab3-4955-4791-9fb8-2641838e5d58"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential_13\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "embedding_12 (Embedding)     (None, 30, 100)           1000000   \n",
      "_________________________________________________________________\n",
      "flatten_12 (Flatten)         (None, 3000)              0         \n",
      "_________________________________________________________________\n",
      "dense_24 (Dense)             (None, 64)                192064    \n",
      "_________________________________________________________________\n",
      "dense_25 (Dense)             (None, 39)                2535      \n",
      "=================================================================\n",
      "Total params: 1,194,599\n",
      "Trainable params: 1,194,599\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "9a3xVAqdKWDF"
   },
   "outputs": [],
   "source": [
    "model.layers[0].set_weights([embedding_matrix])\n",
    "model.layers[0].trainable = False"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "3dFjoEdTKZhW"
   },
   "outputs": [],
   "source": [
    "model.compile(optimizer=\"rmsprop\", loss=\"categorical_crossentropy\", metrics=[\"accuracy\"])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 382
    },
    "colab_type": "code",
    "id": "KdrB6FBKKlj1",
    "outputId": "05aaf7b0-7c9f-4d78-c9ee-de54dc435d6c"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 110630 samples, validate on 27658 samples\n",
      "Epoch 1/10\n",
      "110630/110630 [==============================] - 5s 46us/step - loss: 2.0437 - accuracy: 0.4565 - val_loss: 1.8307 - val_accuracy: 0.4931\n",
      "Epoch 2/10\n",
      "110630/110630 [==============================] - 5s 44us/step - loss: 1.6671 - accuracy: 0.5347 - val_loss: 1.8190 - val_accuracy: 0.5004\n",
      "Epoch 3/10\n",
      "110630/110630 [==============================] - 5s 44us/step - loss: 1.5358 - accuracy: 0.5647 - val_loss: 1.8393 - val_accuracy: 0.4984\n",
      "Epoch 4/10\n",
      "110630/110630 [==============================] - 5s 44us/step - loss: 1.4409 - accuracy: 0.5871 - val_loss: 1.8608 - val_accuracy: 0.4986\n",
      "Epoch 5/10\n",
      "110630/110630 [==============================] - 5s 44us/step - loss: 1.3634 - accuracy: 0.6076 - val_loss: 1.9154 - val_accuracy: 0.4966\n",
      "Epoch 6/10\n",
      "110630/110630 [==============================] - 5s 45us/step - loss: 1.2995 - accuracy: 0.6226 - val_loss: 1.9753 - val_accuracy: 0.4881\n",
      "Epoch 7/10\n",
      "110630/110630 [==============================] - 5s 44us/step - loss: 1.2448 - accuracy: 0.6382 - val_loss: 2.0315 - val_accuracy: 0.4821\n",
      "Epoch 8/10\n",
      "110630/110630 [==============================] - 5s 43us/step - loss: 1.1967 - accuracy: 0.6515 - val_loss: 2.1013 - val_accuracy: 0.4774\n",
      "Epoch 9/10\n",
      "110630/110630 [==============================] - 5s 44us/step - loss: 1.1537 - accuracy: 0.6628 - val_loss: 2.1634 - val_accuracy: 0.4776\n",
      "Epoch 10/10\n",
      "110630/110630 [==============================] - 5s 44us/step - loss: 1.1153 - accuracy: 0.6747 - val_loss: 2.2329 - val_accuracy: 0.4738\n"
     ]
    }
   ],
   "source": [
    "history = model.fit(\n",
    "    x_train,\n",
    "    y_train,\n",
    "    epochs=10,\n",
    "    batch_size=128,\n",
    "    validation_data=(x_val, y_val)\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 52
    },
    "colab_type": "code",
    "id": "6QD_ZRQIKqos",
    "outputId": "95f0f1ae-d65f-407f-be3f-c426291f977c"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "59267/59267 [==============================] - 2s 34us/step\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[2.2618018843262875, 0.4719827175140381]"
      ]
     },
     "execution_count": 161,
     "metadata": {
      "tags": []
     },
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.evaluate(x_test, y_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "ZN6FXYVfJ7YK"
   },
   "outputs": [],
   "source": [
    "# it seems like the combination of Embedding and Dense layer are not able to perform well on text problem.\n",
    "# lets try implementing some advanced architecture which are tuned to work with sequence data."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "tjLkOkpcR6OG"
   },
   "outputs": [],
   "source": [
    "model = Sequential()\n",
    "model.add(Embedding(max_words, 64, input_length=max_len))\n",
    "model.add(SimpleRNN(64))\n",
    "model.add(Dense(n_classes, activation=\"softmax\"))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "rQgXCrj0Up_u"
   },
   "outputs": [],
   "source": [
    "model.compile(\n",
    "    optimizer=\"rmsprop\",\n",
    "    loss=\"categorical_crossentropy\",\n",
    "    metrics=[\"accuracy\"]\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 436
    },
    "colab_type": "code",
    "id": "EKPy6eI0UZBt",
    "outputId": "af70834a-d05a-4bdd-c43b-43f05314ab96"
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/usr/local/lib/python3.6/dist-packages/tensorflow/python/framework/indexed_slices.py:434: UserWarning: Converting sparse IndexedSlices to a dense Tensor of unknown shape. This may consume a large amount of memory.\n",
      "  \"Converting sparse IndexedSlices to a dense Tensor of unknown shape. \"\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 110630 samples, validate on 27658 samples\n",
      "Epoch 1/10\n",
      "110630/110630 [==============================] - 17s 155us/step - loss: 2.7116 - accuracy: 0.3047 - val_loss: 2.3531 - val_accuracy: 0.3759\n",
      "Epoch 2/10\n",
      "110630/110630 [==============================] - 17s 154us/step - loss: 2.1051 - accuracy: 0.4470 - val_loss: 2.0788 - val_accuracy: 0.4609\n",
      "Epoch 3/10\n",
      "110630/110630 [==============================] - 17s 153us/step - loss: 1.8228 - accuracy: 0.5227 - val_loss: 1.8792 - val_accuracy: 0.5101\n",
      "Epoch 4/10\n",
      "110630/110630 [==============================] - 17s 154us/step - loss: 1.6394 - accuracy: 0.5672 - val_loss: 1.8441 - val_accuracy: 0.5239\n",
      "Epoch 5/10\n",
      "110630/110630 [==============================] - 17s 153us/step - loss: 1.5090 - accuracy: 0.5974 - val_loss: 1.7770 - val_accuracy: 0.5434\n",
      "Epoch 6/10\n",
      "110630/110630 [==============================] - 17s 155us/step - loss: 1.4023 - accuracy: 0.6247 - val_loss: 1.7577 - val_accuracy: 0.5415\n",
      "Epoch 7/10\n",
      "110630/110630 [==============================] - 17s 154us/step - loss: 1.3105 - accuracy: 0.6473 - val_loss: 1.8137 - val_accuracy: 0.5331\n",
      "Epoch 8/10\n",
      "110630/110630 [==============================] - 17s 155us/step - loss: 1.2248 - accuracy: 0.6701 - val_loss: 1.8211 - val_accuracy: 0.5350\n",
      "Epoch 9/10\n",
      "110630/110630 [==============================] - 17s 154us/step - loss: 1.1434 - accuracy: 0.6921 - val_loss: 1.8575 - val_accuracy: 0.5300\n",
      "Epoch 10/10\n",
      "110630/110630 [==============================] - 17s 154us/step - loss: 1.0605 - accuracy: 0.7138 - val_loss: 1.8739 - val_accuracy: 0.5298\n"
     ]
    }
   ],
   "source": [
    "history = model.fit(\n",
    "    x_train,\n",
    "    y_train,\n",
    "    epochs=10,\n",
    "    batch_size=128,\n",
    "    validation_data=(x_val, y_val)\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "JriWF0afnKbi"
   },
   "outputs": [],
   "source": [
    "# let us try using LSTM layer\n",
    "\n",
    "model = Sequential()\n",
    "model.add(Embedding(max_words, 64, input_length=max_len))\n",
    "model.add(LSTM(64))\n",
    "model.add(Dense(64, activation=\"relu\"))\n",
    "model.add(Dense(n_classes, activation=\"softmax\"))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 295
    },
    "colab_type": "code",
    "id": "9dTFnxRH9v2q",
    "outputId": "de96dc0a-69a7-4f9e-ee55-05d7e982ea54"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential_1\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "embedding_1 (Embedding)      (None, 30, 64)            640000    \n",
      "_________________________________________________________________\n",
      "lstm_1 (LSTM)                (None, 64)                33024     \n",
      "_________________________________________________________________\n",
      "dense_1 (Dense)              (None, 64)                4160      \n",
      "_________________________________________________________________\n",
      "dense_2 (Dense)              (None, 39)                2535      \n",
      "=================================================================\n",
      "Total params: 679,719\n",
      "Trainable params: 679,719\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "l9uK-WSH9xpw"
   },
   "outputs": [],
   "source": [
    "model.compile(\n",
    "    optimizer=\"rmsprop\",\n",
    "    loss=\"categorical_crossentropy\",\n",
    "    metrics=[\"accuracy\"]\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 454
    },
    "colab_type": "code",
    "id": "rfe05HvZ-Akb",
    "outputId": "a9d80dcb-bd88-4cd5-e9b8-a64b24b70e5b"
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/usr/local/lib/python3.6/dist-packages/tensorflow/python/framework/indexed_slices.py:434: UserWarning: Converting sparse IndexedSlices to a dense Tensor of unknown shape. This may consume a large amount of memory.\n",
      "  \"Converting sparse IndexedSlices to a dense Tensor of unknown shape. \"\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 110630 samples, validate on 27658 samples\n",
      "Epoch 1/10\n",
      "110630/110630 [==============================] - 50s 448us/step - loss: 2.5128 - accuracy: 0.3477 - val_loss: 2.0743 - val_accuracy: 0.4576\n",
      "Epoch 2/10\n",
      "110630/110630 [==============================] - 49s 443us/step - loss: 1.8238 - accuracy: 0.5210 - val_loss: 1.7925 - val_accuracy: 0.5280\n",
      "Epoch 3/10\n",
      "110630/110630 [==============================] - 49s 443us/step - loss: 1.5838 - accuracy: 0.5768 - val_loss: 1.6437 - val_accuracy: 0.5646\n",
      "Epoch 4/10\n",
      "110630/110630 [==============================] - 49s 442us/step - loss: 1.4295 - accuracy: 0.6140 - val_loss: 1.6205 - val_accuracy: 0.5714\n",
      "Epoch 5/10\n",
      "110630/110630 [==============================] - 49s 441us/step - loss: 1.3171 - accuracy: 0.6375 - val_loss: 1.5516 - val_accuracy: 0.5888\n",
      "Epoch 6/10\n",
      "110630/110630 [==============================] - 49s 442us/step - loss: 1.2312 - accuracy: 0.6585 - val_loss: 1.5375 - val_accuracy: 0.5938\n",
      "Epoch 7/10\n",
      "110630/110630 [==============================] - 49s 441us/step - loss: 1.1603 - accuracy: 0.6754 - val_loss: 1.5412 - val_accuracy: 0.5917\n",
      "Epoch 8/10\n",
      "110630/110630 [==============================] - 49s 442us/step - loss: 1.0971 - accuracy: 0.6922 - val_loss: 1.5374 - val_accuracy: 0.5933\n",
      "Epoch 9/10\n",
      "110630/110630 [==============================] - 49s 443us/step - loss: 1.0386 - accuracy: 0.7070 - val_loss: 1.5863 - val_accuracy: 0.5951\n",
      "Epoch 10/10\n",
      "110630/110630 [==============================] - 49s 444us/step - loss: 0.9834 - accuracy: 0.7223 - val_loss: 1.5581 - val_accuracy: 0.5923\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<keras.callbacks.callbacks.History at 0x7f233fde1b70>"
      ]
     },
     "execution_count": 40,
     "metadata": {
      "tags": []
     },
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.fit(\n",
    "    x_train,\n",
    "    y_train,\n",
    "    epochs=10,\n",
    "    batch_size=128,\n",
    "    validation_data=(x_val, y_val)\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 52
    },
    "colab_type": "code",
    "id": "mxtJ4O0nPvH2",
    "outputId": "9aedeab4-164b-427c-d7fc-db5a254137fa"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "59267/59267 [==============================] - 9s 155us/step\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[1.5690422438701446, 0.5901091694831848]"
      ]
     },
     "execution_count": 42,
     "metadata": {
      "tags": []
     },
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.evaluate(x_test, y_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "SYAMbTzL-N7x"
   },
   "outputs": [],
   "source": [
    "model = Sequential()\n",
    "model.add(Embedding(max_words, 64, input_length=max_len))\n",
    "model.add(LSTM(64, return_sequences=True, dropout=0.2, recurrent_dropout=0.2))\n",
    "model.add(LSTM(64, dropout=0.2, recurrent_dropout=0.2))\n",
    "model.add(Dense(64, activation=\"relu\"))\n",
    "model.add(Dense(n_classes, activation=\"softmax\"))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 330
    },
    "colab_type": "code",
    "id": "FLFYpNpkxUP5",
    "outputId": "40be3ada-db66-4b9d-98ab-52aeb03c02f2"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential_4\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "embedding_4 (Embedding)      (None, 30, 64)            640000    \n",
      "_________________________________________________________________\n",
      "lstm_4 (LSTM)                (None, 30, 64)            33024     \n",
      "_________________________________________________________________\n",
      "lstm_5 (LSTM)                (None, 64)                33024     \n",
      "_________________________________________________________________\n",
      "dense_5 (Dense)              (None, 64)                4160      \n",
      "_________________________________________________________________\n",
      "dense_6 (Dense)              (None, 39)                2535      \n",
      "=================================================================\n",
      "Total params: 712,743\n",
      "Trainable params: 712,743\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "b2PQCtGvQGYk"
   },
   "outputs": [],
   "source": [
    "model.compile(\n",
    "    optimizer=\"rmsprop\",\n",
    "    loss=\"categorical_crossentropy\",\n",
    "    metrics=[\"accuracy\"]\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 801
    },
    "colab_type": "code",
    "id": "ByET2TkrQTgN",
    "outputId": "7fe30a75-7a19-4931-c010-cc4387aed01e"
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/usr/local/lib/python3.6/dist-packages/tensorflow/python/framework/indexed_slices.py:434: UserWarning: Converting sparse IndexedSlices to a dense Tensor of unknown shape. This may consume a large amount of memory.\n",
      "  \"Converting sparse IndexedSlices to a dense Tensor of unknown shape. \"\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 110630 samples, validate on 27658 samples\n",
      "Epoch 1/20\n",
      "110630/110630 [==============================] - 93s 843us/step - loss: 2.7107 - accuracy: 0.2952 - val_loss: 2.2446 - val_accuracy: 0.4080\n",
      "Epoch 2/20\n",
      "110630/110630 [==============================] - 92s 836us/step - loss: 2.0610 - accuracy: 0.4615 - val_loss: 1.8930 - val_accuracy: 0.5059\n",
      "Epoch 3/20\n",
      "110630/110630 [==============================] - 92s 828us/step - loss: 1.7843 - accuracy: 0.5316 - val_loss: 1.7260 - val_accuracy: 0.5425\n",
      "Epoch 4/20\n",
      "110630/110630 [==============================] - 92s 832us/step - loss: 1.6261 - accuracy: 0.5662 - val_loss: 1.6257 - val_accuracy: 0.5656\n",
      "Epoch 5/20\n",
      "110630/110630 [==============================] - 92s 833us/step - loss: 1.5140 - accuracy: 0.5930 - val_loss: 1.5663 - val_accuracy: 0.5811\n",
      "Epoch 6/20\n",
      "110630/110630 [==============================] - 92s 832us/step - loss: 1.4313 - accuracy: 0.6119 - val_loss: 1.5145 - val_accuracy: 0.5922\n",
      "Epoch 7/20\n",
      "110630/110630 [==============================] - 92s 830us/step - loss: 1.3729 - accuracy: 0.6242 - val_loss: 1.4820 - val_accuracy: 0.6000\n",
      "Epoch 8/20\n",
      "110630/110630 [==============================] - 92s 829us/step - loss: 1.3252 - accuracy: 0.6350 - val_loss: 1.4768 - val_accuracy: 0.6017\n",
      "Epoch 9/20\n",
      "110630/110630 [==============================] - 92s 828us/step - loss: 1.2875 - accuracy: 0.6433 - val_loss: 1.4578 - val_accuracy: 0.6031\n",
      "Epoch 10/20\n",
      "110630/110630 [==============================] - 91s 826us/step - loss: 1.2546 - accuracy: 0.6524 - val_loss: 1.4584 - val_accuracy: 0.6061\n",
      "Epoch 11/20\n",
      "110630/110630 [==============================] - 92s 829us/step - loss: 1.2267 - accuracy: 0.6581 - val_loss: 1.4454 - val_accuracy: 0.6083\n",
      "Epoch 12/20\n",
      "110630/110630 [==============================] - 92s 831us/step - loss: 1.2027 - accuracy: 0.6642 - val_loss: 1.4515 - val_accuracy: 0.6118\n",
      "Epoch 13/20\n",
      "110630/110630 [==============================] - 92s 828us/step - loss: 1.1791 - accuracy: 0.6688 - val_loss: 1.4439 - val_accuracy: 0.6113\n",
      "Epoch 14/20\n",
      "110630/110630 [==============================] - 92s 829us/step - loss: 1.1586 - accuracy: 0.6729 - val_loss: 1.4520 - val_accuracy: 0.6122\n",
      "Epoch 15/20\n",
      "110630/110630 [==============================] - 92s 829us/step - loss: 1.1392 - accuracy: 0.6784 - val_loss: 1.4430 - val_accuracy: 0.6132\n",
      "Epoch 16/20\n",
      "110630/110630 [==============================] - 92s 830us/step - loss: 1.1209 - accuracy: 0.6831 - val_loss: 1.4511 - val_accuracy: 0.6130\n",
      "Epoch 17/20\n",
      "110630/110630 [==============================] - 92s 830us/step - loss: 1.1031 - accuracy: 0.6872 - val_loss: 1.4510 - val_accuracy: 0.6154\n",
      "Epoch 18/20\n",
      "110630/110630 [==============================] - 92s 830us/step - loss: 1.0882 - accuracy: 0.6905 - val_loss: 1.4581 - val_accuracy: 0.6144\n",
      "Epoch 19/20\n",
      "110630/110630 [==============================] - 92s 831us/step - loss: 1.0726 - accuracy: 0.6932 - val_loss: 1.4680 - val_accuracy: 0.6153\n",
      "Epoch 20/20\n",
      "110630/110630 [==============================] - 91s 825us/step - loss: 1.0593 - accuracy: 0.6982 - val_loss: 1.4618 - val_accuracy: 0.6133\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<keras.callbacks.callbacks.History at 0x7f233e69b128>"
      ]
     },
     "execution_count": 59,
     "metadata": {
      "tags": []
     },
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.fit(\n",
    "    x_train, \n",
    "    y_train, \n",
    "    batch_size=128,\n",
    "    epochs=20,\n",
    "    validation_data=(x_val, y_val)\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 34
    },
    "colab_type": "code",
    "id": "Nw1eGwGcQqDw",
    "outputId": "f276b19a-5181-4182-ddd1-381ee867df95"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "23296/59267 [==========>...................] - ETA: 11s"
     ]
    }
   ],
   "source": [
    "model.evaluate(x_test, y_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "ITBE2IXRgXrF"
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "colab": {
   "name": "NewCategorization.ipynb",
   "provenance": []
  },
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}